GF100’s Gaming Architecture

Although it feels like ages ago, NVIDIA announced the Fermi architecture back in September of 2009, focusing on the compute abilities of the GPU that would be GF100. Today’s announcement is about filling in the blanks – where does the graphics hardware fit in to the design that NVIDIA revealed earlier.

As a quick refresher, the GF100 is composed of 512 SPs, which NVIDIA now calls CUDA cores. Each CUDA core is organized in to a block of 32 cores which we call a Streaming Multiprocessor (SM), which in turn are laid out so that 4 of them are in a Graphics Processing Cluster (GPC). Finally, there are 4 GPCs in a single GF100 chip. 32x4x4 = 512 CUDA cores.

  GF100 GT200x2 (GTX 295) GT200 (GTX 285) G92 (9800+ GTX)
Stream Processors 512 2 x 240 240 128
Texture Address / Filtering 64/256 2 x 80 / 80 80 / 80 64 / 64
ROPs 48 2x 28 32 16
Core Clock ? 576MHz 648MHz 738MHz
Shader Clock ? 1242MHz 1476MHz 1836MHz
Memory Clock ? GDDR5 999MHz (1998MHz data rate) GDDR3 1242MHz (2484MHz data rate) GDDR3 1100MHz (2200MHz data rate) GDDR3
Memory Bus Width 384-bit 2 x 448-bit 512-bit 256-bit
Frame Buffer ? 2 x 896MB 1GB 512MB
Transistor Count 3B 2 x 1.4B 1.4B 754M
Manufacturing Process TSMC 40nm TSMC 55nm TSMC 55nm TSMC 55nm
Price Point $? $500 $400 $150 - 200

When NVIDIA first unveiled GF100, we did not know anything about the ROPs, texture units, or any of the fixed-function graphics units that are customary in a GPU. Today we now have that information, and can discuss where it goes.

While GF100 resembles GT200 in a number of ways as a compute GPU, as a gaming GPU it’s very close to being a complete departure from GT200. The big change here is that the single block of fixed-function hardware as we know it is gone. Virtually every piece of hardware has been split up and moved down a level, and can now be found as part of a GPC or as a SM. NVIDIA’s GF100 block diagram does a good job of showing this.

What’s left of the fixed function hardware is the GigaThread Engine, which is what NVIDIA is calling their scheduler. To understand where our fixed function hardware went, we must meet the newest execution units: The PolyMorph Engine and the Raster Engine.

We’ll start with the Raster Engine. The Raster Engine is the combination of all the raster related hardware that we would have found previously in the fixed function pipeline. It does edge/triangle setup, rasterization, and z-culling in a pipelined manner. Each GPC has its own Raster Engine. NVIDIA has very little to say about the Raster Engine, but very little has changed from GT200 besides the fact that there are now 4 of them (one for each GPC) as opposed to 1 larger block. Each rasterizer can do 8 pixels per clock, for a total of 32 pixels per clock over the entirety of GF100.

The much more important of the two new execution units is the PolyMorph Engine, which is what NVIDIA is calling the execution unit that handles geometry for GF100. The PolyMorph Engine is responsible for handling Vertex Fetch, Tessellation, Viewport Transform, Attribute Setup, and Stream Output. It’s here where the Tessellator is located, one of the biggest changes that DX11 is bringing to GPU design. Unlike the Raster Engine, each SM has a PolyMorph Engine, giving the GF100 16 PolyMorph Engines altogether.

While the PolyMoprh Engine may sound simple in its description, don’t let it fool you. NVIDIA didn’t just move their geometry hardware to a different place, clone it 15 times, and call it a day. This was previously fixed-function hardware where a single unit sat in a pipeline and did its share of the work. By splitting up the fixed-function pipeline like this, NVIDIA in actuality created a lot of work for themselves. Why? Out of order execution.

OoO is something we usually reserve for CPUs, where high-end CPUs are built to execute instructions out of order in order to extract more performance out of them through instruction level parallelism. OoO is very hard to accomplish, because you can only execute certain instructions ahead of other ones while maintaining the correct result for your data. Execute an add instruction that relies on a previous operation before that’s done, and you have problems. GF100 isn’t a full OoO design, so we’re not going to cover OoO in-depth here, but if you’d like to know more please see this article.

At any rate, internally each PolyMorph Engine is still a simple in-order design. NVIDIA hasn’t gone so far as to make a PolyMorph Engine an OoO design – but because there are 16 of them when there used to be just 1, OoO hazards can occur just as they would in an OoO executing CPU. NVIDIA now has to keep track of what each PolyMorph Engine is doing in respect to the other 15, and put the brakes on any of them that get too far ahead in order to maintain the integrity of results.

To resolve the hazards of OoO, GF100 has a private communication channel just for the PolyMorph Engines that allows them to stay together on a task in spite of being spread apart. The fact of the matter is that all of the work that goes in to making a design like this work correctly is an immense amount of effort, and NVIDIA’s engineers are quite proud of this effort. They have taken the monolithic design of prior GPUs’ geometry units, and made it parallel. We can’t overstate how much of an engineering challenge this is.

However all of this work came at a cost, and not just the significant engineering resources NVIDIA threw at GF100. The other cost was time – we believe that the PolyMorph Engine is the single biggest reason that GF100 didn’t make it out last year. It’s the single biggest redesign of any component in GF100, and is something that NVIDIA had to start virtually from scratch on. When NVIDIA told us that designing a big GPU is hard, this is what they had in mind.

Now why did NVIDIA put themselves through all of this? Because in their eyes, they had to. The use of a fixed-function pipeline in their eyes was a poor choice given the geometric complexity that a tessellator would create, and hence the entire pipeline needed to be rebalanced. By moving to the parallel design of the PolyMorph Engine, NVIDIA’s geometry hardware is no longer bound by any limits of the pipelined fixed-function design (such as bottlenecks in one stage of the pipeline), and for better or for worse, they can scale their geometry and raster abilities with the size of the chip. A smaller GF100 derivative will not have as many PolyMorph or Raster units as GF100, and as a result won’t have the same level of performance; G92 derivatives and AMD’s designs both maintain the same fixed function pipeline through all chips, always offering the same level of performance.

Speaking of performance, we’ll get to that in a bit, but for now we’ll finish our discussion of GF100’s graphics units. Each SM has 4 texture units, which have changes somewhat from the GT200. In GT200, the texture units were at the TPC level; here they are per SM. The texture units are also affected by the cache reorganization of GF100; for L1 cache they still have the same 12KB of texture cache per set of 4 texture units, while they all share GF100’s 768KB L2 cache, which is now used by all execution units and not just textures. Each unit can compute 1 texture address and fetch 4 texture samples per clock, a significantly higher texture fetch rate than on the GT200. Furthermore in conjunction with DX11, NVIDIA’s texture units now support DX11’s new compressed texture formats. Finally, texture units are now tied to the shader clock and not the core clock. They run at 1/2 the shader clock speed of GF100.

Last but not least, this brings us to the ROPs. The ROPs have been reorganized, there are now 48 of them in 6 parttions of 8, and a 64bit memory channel serving each partition. The ROPs now share the L2 cache with the rest of GF100, while under GT200 they had their own L2 cache. Each ROP can do 1 regular 32bit pixel per clock, 1 FP16 pixel over 2 clocks, or 1 FP32 pixel over 4 clocks, giving the GF100 the ability to retire 48 regular pixels per clock. The ROPs are clocked together with the L2 cache.

That leaves us on a final note: clocks. The core clock has been virtually done away with on GF100, as almost every unit now operates at or on a fraction of the shader clock. Only the ROPs and L2 cache operate on a different clock, which is best described as what’s left of the core clock. The shader clock now drives the majority of the chip, including the shaders, the texture units, and the new PolyMorph and Raster Engines. Specifically, the texture units, PolyMorph Engine, and Raster Engine all run at 1/2 shader clock (which NVIDIA is tentatively calling the "GPC Clock"), while the L1 cache and the shaders themselves run at the full shader clock. Don’t be surprised if GF100 overclocking is different from GT200 overclocking as a result.

Index Why NVIDIA Is Focused On Geometry
Comments Locked

115 Comments

View All Comments

  • SothemX - Tuesday, March 9, 2010 - link

    WELL.lets just make it simple. I am an advid gamer...I WANT and NEED power and performance. I care only about how well my games play, how good they look, and the impression they leave with me when I am done.
    I own a PS3 and am thrilled they went with Nvidia- (smart move)
    I own and PC that utilizes the 9800GT OC card....getting ready to upgrade to the new GF100 when it releases, last thing that is on my mind is how the market share is, cost is not an issue.

    Hard-Core gaming requires Nvidia. Entry-level baby boomers use ATI.

    Nvidia is just playing with their food....its a vulgar display of power- better architecture, better programming, better gamming.
  • StevoLincolnite - Monday, January 18, 2010 - link

    [quote]So why does NVIDIA want so much geometry performance? Because with tessellation, it allows them to take the same assets from the same games as AMD and generate something that will look better. With more geometry power, NVIDIA can use tessellation and displacement mapping to generate more complex characters, objects, and scenery than AMD can at the same level of performance.[/quote]

    Might I add to that, nVidia's design is essentially "Modular" they can increase and decrease there geometry performance essentially by taking units out, this however will force programmers to program for the lowest common denominator, whilst AMD's iteration of the technology is the same across the board, so essentially you can have identical geometry regardless of the chip.
  • Yojimbo - Monday, January 18, 2010 - link

    just say the minimum, not the lowest common denominator. it may look fancy bit it doesn't seem to fit.
  • chizow - Monday, January 18, 2010 - link

    The real distinction here is that Nvidia's revamp of fixed-function geometry units to a programmable, scalable, and parallel Polymorph engine means their implementation won't be limited to acceleration of Tesselation in games. Their improvements will benefit every game ever made that benefits from increased geometry performance. I know people around here hate to claim "winners" and "losers" around here when AMD isn't winning, but I think its pretty obvious Nvidia's design and implementation is the better one.

    Fully programmable vs. fixed-function, as long as the fully programmable option is at least as fast is always going to be the better solution. Just look at the evolution of the GPU from mostly fixed-function hardware to what it is today with GF100...a fully programmable, highly parallel, compute powerhouse.
  • mcnabney - Monday, January 18, 2010 - link

    If Fermi was a winner Nvidia would have had samples out to be benchmarked by Anand and others a long time ago.

    Fermi is designed for GPGPU with gaming secondary. Goody for them. They can probably do a lot of great things and make good money in that sector. But I don't know about gaming. Based upon the info that has gotten out and the fact that reality hasn't appeared yet I am guessing that Fermi will only be slightly faster than 5870 and Nvidia doesn't want to show their hand and let AMD respond. Remember, AMD is finishing up the next generation right now - so Fermi will likely compete against Northern Isles on AMDs 32nm process in the Fall.
  • dragonsqrrl - Monday, February 15, 2010 - link

    Firstly, did you not read this article? The gf100 delay was due in large part to the new architecture they developed, and architectural shift ATI will eventually have to make if they wish to remain competitive. In other words, similarly to the g80 enabling GPU computing features/unified shaders for the first time on the PC, Nvidia invested huge resources in r&d and as a result had a next generation, revolutionary GPU before ATI.

    Secondly, Nvidia never meant to place gaming second to GPU computing, as much as you ATI fanboys would like to troll about this subject. What they're trying to do is bring GPU computing up to the level GPU gaming is already at (in terms of accessibility, reliability, and performance). The research they're doing in this field could revolutionize research into many fields outside of gaming, including medicine, astronomy, and 'yes' film production (something I happen to deal with a LOT) while revolutionizing gaming performance and feature sets as well

    Thirdly, I would be AMAZED if AMD can come out with their new architecture (their first since the hd2900) by the 3rd quarter of this year, and on the 32nm process. I just can't see them pushing GPU technology forward in the same way Nvidia has given their new business model (smaller GPUs, less focus on GPU computing), while meeting that tight deadline.
  • chewietobbacca - Monday, January 18, 2010 - link

    "Winning" the generation? What really matters?

    The bottom line, that's what. I'm sure Nvidia liked winning the generation - I'm sure they would have loved it even more if they didn't lose market share and potential profits from the fight...
  • realneil - Monday, January 25, 2010 - link

    winning the generation is a non-prize if the mainstream buyer can only wish they had one. Make this kind of performance affordable and then you'll impress me.
  • chizow - Monday, January 18, 2010 - link

    Yes and the bottom line showed Nvidia turning a profit despite not having the fastest part on the market.

    Again, my point about G80'ing the market was more a reference to them revolutionizing GPU design again rather than simply doubling transistors and functional units or increasing clockspeeds based on past designs.

    The other poster brought up performance at any given point in time, I was simply pointing out a fact being first or second to market doesn't really matter as long as you win the generation, which Nvidia has done for the last few generations since G80 and will again once GF100 launches.
  • sc3252 - Monday, January 18, 2010 - link

    Yikes, if it is more than the original GTX 280 I would expect some loud cards. When I saw those benchmarks of farcrry 2 I was disappointed that I didn't wait, but now that it is using more than a GTX 280 I think I may have made the right choice. While right now I wan't as much performance as possible eventually my 5850 will go into a secondary pc(why I picked 5850) with a lesser power supply. I don't want to have to buy a bigger power supply just because a friend might come over and play once a week.

Log in

Don't have an account? Sign up now